The document provides a survey of research on sensor association rules for mining behavioral patterns from wireless sensor network data. Sensor association rules aim to discover temporal relationships between sensor nodes by detecting correlated events. Various approaches are discussed, including techniques for distributed in-network mining, handling data streams, reducing redundancy, and applying association rules to applications like missing data estimation. Overall, the survey finds that sensor association rules are an effective knowledge discovery technique for wireless sensor networks.

1. International Journal of Engineering Research and Development
e-ISSN: 2278-067X, p-ISSN: 2278-800X, www.ijerd.com
Volume 9, Issue 9 (January 2014), PP. 67-71
Sensor Association Rules: A survey
Prachi singh
Department of computer science, MITS Gwalior, Madhya Pradesh, India
Abstract:- Recently, knowledge Discovery Process has proven to be a promising tool for extracting behavioral
patterns regarding sensor nodes from wireless sensor networks. This paper presents a review of the available
literature on the sensor association rules to find behavioral patterns between the sensor data.
Keywords:- Wireless Sensor Networks, Association Rule Mining
I.
INTRODUCTION
Wireless Sensor Networks (WSNs) have been successfully applied for detailed observation of a variety
of real-world applications, especially in battle fields, smart buildings, the human body, and the other
applications that require the fine-grain monitoring of physical environments that are subject to critical
conditions, such as fires, toxic gas leaks, and explosions. With such applications come new challenges for
information processing in sensor networks. Unlike centralized system a sensor network is subject to a unique set
of resource constraints such as finite on board battery power and limited network communication bandwidth. In
a typical sensor network, each sensor node operates and has a microprocessor and a small amount of memory
for signal processing and task scheduling. So, WSN suffers from lot of problems such as lost messages, delay in
data delivery, loss of data, and data redundancy, etc., resulting in poor quality of service (QoS). Various
techniques and protocols have been suggested to improve the QoS of WSN.
Recently , the knowledge discovery process, which is a well known process in traditional database
systems used to extract patterns from data, has shown to be a promising tool to improve WSN performance and
its QoS. Knowledge discovery in WSN (KDW) has been used to extract the following two types of information
(knowledge): first is, patterns about the surrounding environment, which are extracted from the data reported by
sensor nodes, and the second is, behavioral patterns about sensor nodes, which are extracted from meta-data
describing sensor‟s behaviors.Many techniques have been proposed to solve the mining frequent patterns
problem from the databases, these techniques differ mainly in the way they represent the database and in which
they generate the frequent patterns. These can be classified into two main approaches: the candidate generation
approach and the pattern growth approach. The candidate generation approach enumerates the frequent patterns
in a level wise manner, with several scans of database. At each level, the patterns found to be frequent are used
to generate the candidates for the next level. Within this approach are the Apriori, Directed Hashing & Pruning,
the partitioning algorithm, and Dynamic Itemset Counting etc. A comprehensive survey of recent algorithms for
mining association rules is presented in [18]. The common algorithm of candidate generation approach is the
Apriori. This algorithm first introduced by Agrawal et al.,[1] in 1993 to discover the correlations between
objects in transactional databases. This algorithm was able to predict the items that can be purchased within the
same transaction. These predictions have a great impact on making decisions about which item should be put on
sale or which items should be placed near to each other.
Association Rule Mining is a costly process. The complexity increases exponentially with the number
of the items presented in the database.
In [2], Loo et al., proposes a framework for extracting association rules from sensor networks. The
association involves the sensors‟ values as the main objects in the rule. In their methodology, the time is divided
into intervals, and the sensors‟ values at that interval formulate the transaction to be stored in the database. Each
different value of a sensor is regarded as single element and it is assumed that sensors take on a finite number of
discrete states. Each transaction is associated with a weight value indicating the validity of the transaction (i.e.
the duration of time in which there were no changes in sensors‟ values). The support of the pattern is then
defined to be the total length of all non overlapping intervals in which the pattern occurs. The main differences
between this methodology and the sensor association rules proposed in [5], is that in sensor association rules,
sensors themselves act as the main objects, not their values, and the support of the pattern is the number of
epochs in which the pattern occurs as a subset. The algorithm presented in his work is complex and to handle the
streamed data is also difficult and time consuming. Most of the data mining techniques applied to sensor data
are based on centralized data extraction, where the data is collected at a single site and then the mining
technique is applied. However, there are still some techniques that use the distributed nature of sensors and try
to apply distributed mining algorithms.
67

2. Sensor Association Rules: A survey
In [3], Key Romer et al, proposes an in-network data mining technique to discover frequent patterns of
events with certain spatial and temporal properties. In their framework, each sensor should be aware of the
events that are within a certain distance from itself (this distance may be a Euclidian distance or number of
hops). The sensor then collects these events and applies a mining algorithm to discover the pattern that satisfies
a given parameters. The main difference between Romer‟s methodology and the sensor association rule
proposed in [5], is in type of events in the discovered pattern, in which he targeted all the possible values of a
sensor as objects, whereas in [5] targets sensors in an abstract manner regardless of the sensor value, which will
reduce the amount of the data exchanged between the nodes if we assume in-network solution.
In [4] Miahail and Le, propose a data model to store and update the data generated from sensor
networks. In their work, they adopt a centralized extraction methodology where all the data is collected at sink
for further analysis. They focus on generating association rules between pairs of sensors instead of generating all
associations. The direct application of their extracted rules is to determine the set of the sensors to be used to
predict a missed value of other sensors.
II.
SENSOR ASSOCIATION RULES
The concept of sensor association rules (SAR) first introduced by Azzedine Boukerche and Samer
Samarah [5]. In their work they introduce a new formulation for the association rules, which is able to generate
the time relations between sensor devices in a particular sensor network. In this formulation they allow
traditional data mining algorithms proposed to solve the classical association rule mining problem to be applied
on sensor based class of applications that generate and use sensor data. The generated rules will give a clear
picture about the correlations between sensors in the network and can be used to predict the sources of the future
events. In their work they also suggested two possible methodologies for extracting the data from wireless
sensor networks. The first is a direct transmission, in which the data is transmitted to the central site without any
optimization from the nodes. The second method considers the overall limited resources in the network and each
node tries to optimize the number of messages it will send.
Sensor association rules is among the first knowledge discovery techniques that have been proposed to
generate behavioral patterns. It discovers the temporal relationships between sensor nodes in detecting events.
An example of sensor association rules could be (A→B, λ, 60%). Here A and B are the sets of the sensors. The
means of this rule is that if sensors from A detect events within time λ, then there is 60% chance that sensors
from B detects events within that same time interval. The main benefit of Sensor Association Rules (SAR) in
many applications is the ability to predict sources of future events, and the ability to identify the sets of
temporally correlated sensors.
Definition: Let A= {s1,s2,s3…….,sm} be a set of sensors in a particular sensor network. We assume
that the time is divided into equal sized slots {t1,t2,…..,tn} such that ti+1 - ti = λ, for all 1 < i < n. λ is the size of
each time slot. This = tn - t1 is the historical period of the data defined during the data extraction process. A set P
= {s1,s2,…..,sk} ⊆ A is called a pattern of sensors.
A sensor database, DS, is defined to be a set of epochs in which each epoch is a couple E(E ts,P) such that P is a
pattern of sensors that report events within the same time slot. E ts is the epoch‟s time slot. We say an epoch
E(Ets, P) supports a pattern P1 if P1 ⊆ P. frequency of the pattern P1 in DS is defined to be the number of epochs
in DS that support it.
Freq (P1, DS) = | {E(Ets, P)| P1 ⊆ P} |.
Sensor association rules are implications of the form P‟→P‟‟ where P‟⊂A, P‟‟⊂A and P‟∩P‟‟ = ⏀. The
frequency of the rule (P‟→P‟‟) is the frequency of the pattern (P‟∪P‟‟). The confidence of the rule is defined as
follows:
Conf(P‟→P‟‟) = Freq(P‟∪P‟‟, DS) / Freq(P‟, DS)
The major impacts of Sensor Association Rules that could bring benefit to many applications are the ability to
predict sources of future events, and the ability to identify sets of temporally correlated sensors.
Le Gruenwald et al, [7] proposed a new approach to find associations between the sensors in any sensor
network. They incorporate the time factor to estimate the missing sensor values. They propose a data estimation
technique, FARM, which uses association rule mining to discover intrinsic relationships among sensors and
incorporate them into the data estimation while taking data freshness into consideration. FARM uses association
rule mining to find related sensors. The contribution of FARM is:

Incorporate the temporal aspect into association rules and estimation

Compact data streams and allow a large history to appropriately influence sensor rules

Guarantee retrievability of original data from its compact form
Azzedine Boukerche et al, proposed in [5], a new representation structure for mining
association rules from wireless sensor networks that is a tree structure which is known as the PLT (Positional
68

3. Sensor Association Rules: A survey
Lexicographic Tree), a better solution than the FP-tree proposed in [], to find the frequent sensor patterns from
the set of epochs (DS). The mechanism proposed in his work uses a partitioning mechanism in PLT, that is used
to locate the conditional vectors of a particular pattern easily, instead of following the node‟s link as in the FPtree. These partitions are independent hence, we don‟t need the entire structure to be in the main memory, as
opposed to the FP-tree that requires the whole structure to be in the main memory. It also uses a numerical value
that is known as the comparison value of the position vectors, is used to locate the insert vectors accelerate the
process of accessing the PLT structure, whereas the FP-tree doesn‟t include such values. PLT requires smaller
variable sizes for storing the positional values, since it uses the lexicographic distance to represent sensors, as
opposed to the FP-tree that uses the actual sensor‟s identifier.
In [9], Samer Samarah et al, proposed coverage based association rules for wireless vehicular ad hoc
networks, to discover the correlation among the set of locations monitored by the network. Coverage based rules
have been designed specifically for sensor networks that guarantee a k-coverage property for the area under
monitoring. Here k-coverage property states that in a k-coverage sensor network, all k nodes within a specific
area are expected to detect the same event, as a result all k sensor nodes will have the same activity sets during
the process of profiling their behaviors when preparing the data needed for generating sensor association
patterns.. Author proposed a relaxation version of sensor association rules that discovers the correlation between
a set of locations (areas) rather than individual sensor nodes.
Sayed Khairuzzaman Tanbeer et al in [10] proposes a new tree-based data structure called sensor
pattern tree (SP-tree) to generate association rules from WSNs data with one database scan. This SP-tree is able
to capture the information with one scan over the stream of sensor data and store them in memory-efficient
highly compact manner, similar to FP-tree. The main idea of SP-tree is to obtain the frequency of all eventdetecting sensors‟ data and construct a prefix-tree based on that in any canonical order, then reorganize the tree
in a frequency descending order. Once the SP-tree is constructed, we apply the efficient FP-growth mining
technique on it. The basic operations in the FP-growth based pattern growth mining approach are counting the
set of frequent event-detecting sensors, constructing conditional pattern base for each of such sensor, and
constructing new conditional tree from each conditional pattern base. [19]
In [11], Azzedine et al, proposed an in network data reduction mechanism to reduce the amount of data
(about sensor‟s behaviors) by removing some of the data‟s redundancies. The proposed in network data
reduction is implemented on top of a data gathering tree, which they refer to as a minimum nodes data gathering
tree (MNDGT) and in cooperation with the distributed extraction mechanism. The MNDGT takes in
consideration that not all sensor nodes are going to participate in formulating the sensor association rules, and it
is constructed in such a way that those nodes that will participate in formulating sensor association rules are
included, plus the nodes that are needed to maintain the minimum distance to the sink. The MNDGT is designed
in such a way that the activity sets of different nodes can meet at intermediate nodes where a reduction
technique can take place.
In [16], Anjan Das, proposes an enhanced data reduction mechanism to gather data for mining sensor
association rules. In this the he tried to enhance the algorithm proposed by Azzedine et al in [11], by removing
more redundancy between sensor activities.
In [12], Samer et al, proposed target based association rules for point of coverage wireless sensor
networks. They propose a new type of behavioral patterns, which they refer to as target based association rules,
to discover the correlation among a set of targets monitored by a network at border region. Such association can
be the main driver for predicting the source of future events. The main difference between TAR and SAR is that
in TAR the interest will be in capturing the association between the targets, instead of the sensor nodes. He also
proposes two mechanisms to find the frequent target patterns. First is the all-nodes based data preparation
mechanism which is similar to the direct reporting mechanism, requires each sensor node to be active all the
time, without a predefined schedule. Although this solution sounds simple and does not put extra load in the
network, it is costly solution in terms of energy consumption. Second is the schedule-buffer based data
preparation mechanism, which solves the problems in the previous mechanism. A global schedule has been
defined by the sink. The global schedule defines the activity time for each sensor and the set of targets that
should be monitored by each sensor. This solves the energy consumption problem because not all the sensors
have to be active all the time. But to follow the global time schedule each sensor node in same schedule should
have to synchronize with each other to generate the activity set.
In [15], author proposes one more mechanism that is the fused-schedule-buffer based data preparation
mechanism, which is an improved version of the Schedule-Buffer based mechanism. The improvement consists
of allowing the sensor nodes to collaborate with each other during the data preparation process. Although it
reduces the redundancy in the messages to be sent to the sink but it requires much load on individual sensor.
In [14], Nan Jiang et al, proposed a post mining of non redundant association rules for sensor
estimation. They focus on post mining of non-redundant and informative association rules that match the user
interests. The generated association rules are then applied to sensor network databases of a traffic monitoring
69

4. Sensor Association Rules: A survey
site for missing data estimation purpose, in which data missing by a sensor a estimated using the data generated
by its related sensors. Author proposed an algorithm that is developed based on non-redundant informative
association rules which means all rules cannot be derived from other rules which means all rules cannot be
derived from other rules and the left hand side and right hand side of the selected rules contain the input itemset.
In [13], JunTan et al, proposed a Fd-tree method which requires no scanning of the whole data stream
and to only scan the updated transactions once without involving candidate sets generation. A data stream is na
ordered sequence of items that arrives in timely order. Traditional association rules mining algorithms are
developed to work on static data, therefore, cannot be applied directly in data streams. For example, both the
Apriori and the FP-tree mining [19] approaches belong to batch mining. That is, they must process all the
transactions in a batch way. In data stream, new transaction insertion and old transaction deletions may lead to
previously discovered association rules no longer interesting, and new interesting association rules may also
appear. To solve this problem author proposes a new method that uses a Fd-tree structure to store all the
information and requires (1) no scanning of the original database, (2) to only scan the updated transactions once
without involving candidate sets generation.
TABLE I
S. no
Author
Year Key Approach
Association
Applications
between
M. Halatchev
et al
Key Romer et
al
Boukerche et
al
2005
Centralized
extraction
Spatial & temporal
features of events
Positional
Lexicographic Tree
(PLT)
Freshness
Association Rule
Mining (FARM)
Distributed
extraction
Pairs of sensors
[7]
Gruenwald et
al
2007
[8]
Boukerche
al
2007
[9]
Samarah et al
2008
Coverage
based
association rules
Locations
[10]
[11]
Tanbeer et al
Boukerche et
al
2009
2009
sensors
sensors
[12]
Samarah et al
2009
[13]
Jun Tan et al
2010
SP-tree
Minimum
node
data gathering tree
(MNDGT)
Target Association
Rules
Fd-tree
[16]
Anjan Das
2012
In-network
mechanism
sensors
[4]
[3]
[5]
et
2006
2007
Sensor‟s values
Sensors
Estimate the missing sensor
values
Estimate the missing sensor
values
Predict the source of the
future events
Sensor‟s values
Estimation
of
sensor values
Sensors
Estimation
of
missing
sensor values, source of the
future events, fault detection
Predict the location of the
future events in vehicular
adhoc & sensor network
based applications
Sensor association rules
Sensor association rules
Targets
Sensor‟s values
missing
Predict the source of the
future events
Estimation
of
missing
sensor values
Estimation
of
missing
sensor values
In [17], Anjan Das proposes a novel association rule mining mechanism in wireless sensor networks. In
this he proposes an in network mechanism to find frequent sensor patterns in the sensors themselves. So, the
sensors send only the frequent sensor patterns to the sink, not the sensor‟s complete activity sets. He proposes a
mechanism to calculate the frequent sensor patterns in the sensor patterns along with support to the sink.
Although this work reduces redundancies in the messages to be sent to the sink, it put extra load on the
individual sensor node and it requires extra memory in each sensor node to store the information to calculate the
frequent sensor
patterns in the sensors themselves. Table I highlights the key features proposed for WSN in different
works discussed in this paper.
70

5. Sensor Association Rules: A survey
III.
CONCLUSION & FUTURE WORK
In this paper we discussed sensor association rule and its different variations. There are still many
challenges that need to be considering in sensor association rules like many of its applications based on
estimation of missing values and predict the source of the future events, the energy efficiency still a very crucial
issue to apply these techniques on tiny sensor nodes which have the small memory and processing power.
Future research scope in sensor association rules is to enhance the efficiency of wireless sensor network with
respect to energy constraint.
REFERENCES
[1].
[2].
[3].
[4].
[5].
[6].
[7].
[8].
[9].
[10].
[11].
[12].
[13].
[14].
[15].
[16].
[17].
[18].
[19].
R. Agrawal, T. Imielinski, and A. N. Swami, “Mining Association Rules Between Sets of Items in
Large Databases”, Proc. ACM SIGMOD conference on management of Data, 1993, pp. 207-216.
K.K. Loo, L Tong, B. Kao, and D. Chenung, “Online Algorithms for Mining Inter-Stream Association
from Large Sensor Networks,” Proc. Ninth Pacific-Asia Conf. Knowledge Discovery and data Mining
(PAKDD „05), may 2005.
K. Romer, “Distributed Mining of Spatio-temporal Event Patterns in Sensor Networks,” proc. EuroAmerican Workshop Middleware for Sensor Networks (EAWMS „06) June 2006.
M. Halatchev and L. Gruenwald, “Estimating Missing Values in Related Sensor Data Streams,” Proc.
11th Int‟l Conf. Management of Data (COMAD ‟05), Jan. 2005.
Azzedine Boukerche and Samer Samarah, “A New Representation Structure for Mining Association
Rules from Wireless Sensor Networks”, PARADISE University of Ottawa, Canada, 2007
A Boukerche, Samer Samarah, “A novel algorithm for mining association rules in wireless ad hoc
sensor networks”, IEEE Transactions on Parallel and Distributed Systems, Vol 19, No 7, July 2008, pp
865-877.
Le Gruenwald, Hamed Chok, Mazen Aboukhamis, “Using Data Mining to Estimate Missing Sensor
Data”, Seventh IEEE International Conference in Data Mining- Workshops, 2007
Azzedine Boukerche and Samer Samarah, “An Efficient data Extraction Mechanism for Mining
Association Rules from Wireless Sensor Networks”, 2007.
Samer samara, Azzedine Boukerche, and Yonglin ren, “Coverage-based Sensor Association Rules for
Wireless Vehicular Ad hoc and Sensor Networks”, PARADISE Research Laboratory – University of
Ottawa – Canada, 2008.
Syed Khairuzzaman Tanbeer, Vhowdhury Farhan Ahmed, Byeong-Soo Jeong, Young-Koo Lee,
“Efficient Mining of Association Rules from Wireless Sensor Networks”, Feb. 15-18,2009 ICACT
2009.
Azzedine Boukerche and Samer Samarah. “ In-Network Data Reduction and Coverage-Based
Mechanisms for Generating Association Rules in Wireless Sensor Networks”, IEEE Transactions on
Vehicular Technology, Vol. 58, No. 8, Oct. 2009.
Samer Samarah, Alexandre Shema Habyalimana, Azzedine Boukerche, “ target-based Association
Rules for Point-of-Coverage Wireless Sensor Networks”, PARADISE research Laboratory- SITE,
2009.
Jun Tan, Yingyong Bu and haiming Zhao, “Incremental Maintenance of Association Rules Over Data
Streams “, 2010 International Conference on Networking and Digital Society.
Nan Jiang, Zhiqiang Chen, “Post Mining of Non-redundant Association Rules for Sensor Data
Estimation”, 2010 2nd International Conference on Education Technology and Computer (ICETC).
Samer Samarah, Azzedine Boukerche, “target Association Rules: A new Behavioral Patterns for Point
of Coverage Wireless Sensor Networks”, IEEE Transactions on Computers, Vol. 60, No. 6, June 2011.
Anjan Das, “An Enhanced Data Reduction Mechanism to Gather Data for Mining Sensor Association
Rules”, IEEE, 2011.
Anjan Das, “A Novel Association Rule Mining Mechanism in Wireless Sensor Networks”, IEEE,
2012.
B. Goethals, “Survey on Frequent Pattern Mining”, 2003.
J. Han, J. Pei, and Y. Yin, “Mining Frequent Patterns Without Candidate Generation”, Proc. The 2000
ACM SIGMOD INT. Conference on Management of Data, 2000, pp. 1-12.
71